Computer Science > Computation and Language
[Submitted on 26 Jan 2023 (v1), last revised 23 Jul 2023 (this version, v2)]
Title:DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature
View PDFAbstract:The increasing fluency and widespread usage of large language models (LLMs) highlight the desirability of corresponding tools aiding detection of LLM-generated text. In this paper, we identify a property of the structure of an LLM's probability function that is useful for such detection. Specifically, we demonstrate that text sampled from an LLM tends to occupy negative curvature regions of the model's log probability function. Leveraging this observation, we then define a new curvature-based criterion for judging if a passage is generated from a given LLM. This approach, which we call DetectGPT, does not require training a separate classifier, collecting a dataset of real or generated passages, or explicitly watermarking generated text. It uses only log probabilities computed by the model of interest and random perturbations of the passage from another generic pre-trained language model (e.g., T5). We find DetectGPT is more discriminative than existing zero-shot methods for model sample detection, notably improving detection of fake news articles generated by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline to 0.95 AUROC for DetectGPT. See this https URL for code, data, and other project information.
Submission history
From: Eric A Mitchell [view email][v1] Thu, 26 Jan 2023 18:44:06 UTC (3,041 KB)
[v2] Sun, 23 Jul 2023 04:18:36 UTC (1,229 KB)
References & Citations
export BibTeX citation
Loading...
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)